Table 1 Related work

Shin [8]

Distributed Intelligent Routing

Utilize topology information effectively to improve routing decisions

GNN-based algorithm

Prone to causing loops, requires powerful model computing capabilities, and may need modifications to routing protocols

Rischke [14]

Reinforcement Learning Routing

Optimize network load balancing, latency, and packet loss

RSIR algorithm using Q-learning

Limited perception capabilities and potential performance issues

Wang [16]

Deep Reinforcement Learning

Optimize throughput, latency, and packet loss for mouse and elephant flows in data center networks

DQN-based routing policy using deep neural networks and reinforcement learning

Computational complexity limits hardware deployment

Bernárdez [18]

Multiagent Reinforcement Learning

Minimize network congestion

Combination of MARL and GNN

Computational complexity may limit hardware deployment

Chen [19]

Ensemble Learning DRL

Maximize the utilization of optical transport networks

DRL intelligent routing algorithm based on ensemble learning and information propagation neural networks

Computational complexity may limit hardware deployment

Rusek [24]

Deep Learning-Assisted Routing

Establish relationships between network status, topology, traffic matrices, and routing path models

GNN and LSTM models combined with heuristic algorithms

Process of replacing traditional routing algorithms with deep learning-based ones is challenging due to non-convex loss functions, gradient issues, and practical deployment constraints

Farshin [26]

Knowledge-Based Metaheuristics

Utilize knowledge from SDN controllers for VNF placement and routing

Enhanced ant colony system algorithm with knowledge integration

Neglects network volatility and complexity, instability in algorithm training and convergence

Samarji [27]

Fault tolerance metaheuristic

Maximize the network connectivity, maximize the load balance among controllers, minimize the worst-case latency, and maximize the network lifetime

Genetic algorithm and greedy randomized adaptive search problem algorithm

The impact of load distribution of faulty controller on the network performanc is not analyzed

Samarji [28]

Energy soaring-based routing

Selects the network cluster heads for solving the controller placement problem

Energysoaring routing algorithm adopted from the albatross bird

Without factoring in the network's instability and intricacies

Raouf [29]

Ant Colony Optimization

Handle dynamic network fluctuations and reduce congestion, latency, and packet loss

ACOSDN algorithm using Ant Colony Optimization

Sluggish convergence and potential convergence to local optima

Isyaku [30]

Route Path Selection Optimization

Elevate data throughput and packet delivery rates with link quality estimation and constraint parameters

Route path selection optimization approach based on link quality estimation and switch awareness

Doesn't adopt a global optimization perspective, may limit network performance